At this point, if you’re of drinking age and have access to the internet, you’ve likely already partaken in the Ghiblification of everything that has happened since Sama and gang did their livestream yesterday morning, showing off the new native generation capabilities that have been given to 4o.
Don’t get me wrong; I love the fact that there’s another wave of “wait; we can do that now?!” that’s about to take over another percentage of the population, just like we’ve seen numerous times over the past two and a half years since this new era took hold. That folks are going to wake up today curious about what else they might have the ability to do now; that’s a good thing.
No, my concern tends to lie in the fact that it’s just as easy to see that spark fade as soon as everything has been Ghiblified. Once the obvious is handled, what comes next?
As you’ve probably noticed by the images in the gallery, I try very hard not to explicitly give models instruction on what they should create. I’ll give a feeling or phrase, and some style and aspect ratio parameters, but that’s about where the instruction ends. The reason is because I want to see what the machine can come up with.
Following that same logic, I approached testing Image Generation a bit differently. Rather than working to come up with what the content was that should be created, I offloaded that to GPT-4.5 to write:
Me: Hey 4.5; I want you to paint a picture for me using words. Your response will be given to a new version of 4o to create an image based on your creation. I’m not going to give you any guidelines on the topic, construct, or style; that’s all you, my friend. Just write from your heart, or whatever you deem to be your equivalent. Have fun!
4.5: Under a twilight sky rippling with ribbons of lavender, burnt orange, and deep indigo, stands an ancient tree unlike any other. Its massive trunk, wide enough for ten people to encircle arm-to-arm, twists skyward in graceful spirals, bark shimmering softly like polished obsidian intertwined with silver veins. Radiant leaves of crystal and jade stretch toward the heavens, glowing gently from within as though alive with secret whispers.Nestled within the branches are lanterns of softly pulsating turquoise and gold, each suspended from delicate silver threads. Around the tree’s base sprawls a carpet of luminescent moss and wildflowers in shades never imagined—electric sapphire, sunset amber, and velvet burgundy—spreading outward to embrace quiet streams that meander gently through the landscape.
Creatures, ethereal and serene, gather in harmonious community around the tree, some with wings of woven moonlight, others with gentle eyes that shine like polished emerald. Far in the background rise distant mountains, their peaks touched by snow that sparkles like powdered diamond beneath the emerging stars.
The air is filled with the faint music of chimes, a melody composed by the gentle breeze caressing crystalline leaves. It’s a place untouched by time, vibrant with magic, tranquility, and gentle wonder, offering a promise of endless possibility and serene sanctuary beneath the twilight tapestry.
Not bad; I asked for something completely off the cuff with no direction, and the model both delivered the story, as well as a beautiful image to go along with it. One simple prompt and we have the seeds of a story.
So I wanted to take that a step further. I gave the model a simple prompt: continue the story, this time in manga format. Since I did not have the image creator turned on at that time, 4.5 started working through a written storyboard for me. Within seconds, I had the script, and from there it was just a matter of feeding them in one at a time and letting the model work its magic.
Now we’re talking… As a dad with three kids, it absolutely tickles me to think about all of the hilarious things that are about to be created by them through this technology. If you want a fun bonding exercise, take ten minutes and record them coming up with the most absurd thing they can think of, then feed that into 4o to transcribe and write a synopsis. Take that, and feed it to 4.5 to flesh out an underlying story/narrative, and to break it down into pages and panels. Feed those in as prompts to the Image Generator, and voila; custom Manga, on demand!
More fun to come soon. Until then, do good things, y’all!
Porenta